集水區總量管制之不確定性分析研究-定性與定量不確定性分析應用

Abstract

本研究目的在評估集水區總量管制制度(Total Maximum Daily Loads, TMDL)中不確定性的影響，並建立一個整合性分析架構，包含定性與定量的不確定性分析。總量管制以水體涵容能力作為污染源管制基礎，其概念廣泛應用於全世界的集水區管理，其中，不確定性的影響是相當重要的議題並應包含在TMDL計畫規劃當中，本研究即針對此問題進行深入分析探討。 不確定性分析的重要性在各種領域中早已被注意，然而不確定性分析方法卻不如它的重要性一樣，也受到廣泛的應用。不確定性研究大多著重在可量化的不確定性(quantifiable uncertainty)上，不可量化的不確定性(non-quantifiable uncertainty)則受限於主觀特性以及分析方法而被忽略。現有的不確定性分析方法缺乏對於兩種不確定性分析的整合性架構，因此本研究建立一套不確定性分析方法，包括定性不確定性分析(qualitative uncertainty analysis, QLUA)與定量不確定性分析(quantitative uncertainty analysis, QTUA)，並嘗試利用不確定性診斷圖(diagnose figure)以及不確定性價值函數(uncertainty value function, UV)將這兩種分析結果整合。QLUA用於檢視整體系統的不確定性程度，以信任度(confidence)表示，而QTUA則量化不確定性對於TMDL污染分配的影響，以變異度(variability)代表。在QLUA方法中，TMDL規劃所遭遇到的抽象且主觀的不確定性藉由定性檢視表(qualitative check list)、信任函數(belief function)以及專家評量( expert elicitation)方法得之，此信任程度代表該TMDL規劃的不確定性程度，易言之，即決策品質(decision quality)。QTUA特別針對模式使用參數的不確定性對於最佳規劃的影響進行分析，最後得到TMDL分佈曲線作為決策參考，其中模擬模式參數以及設計流量的不確定性分別以蒙地卡羅(Monte Carlo)法以及年流量延時法(Annual Flow Duration Curve)分析。最後建立不確定性診斷圖來判斷其決策品質，若需進一步降低不確定性，則以不確定性價值來量化不確定性降低對決策影響的價值。透過UV計算，決策者可初步了解不確定性降低所需的成本，以及不確定性分析對決策的實際價值，因此決定不確定性降低的程度。本研究以高屏溪以及大漢溪作為QLUA以及QTUA的案例說明。 決策過程中不可避免會遭遇到不確定性問題，透過對不確定性影響的了解以及掌握，有助於決策品質提升，減少決策失敗率。本研究提供一個整合性分析架構，從不確定性的來源定義、分析方法、分析結果以及不確定性降低的價值等，幫助集水區總量管制在未來執行上，可以適當處理不確定性以及其影響，改善TMDL規劃的決策品質。

This dissertation is a management-oriented study, evaluating uncertainty effects on TMDL (Total Maximum Daily Loads) decisions. The TMDL program is a water quality management with regular standard process and is verified and applied widely, especially in the U.S. However, uncertainty problems are occurred inevitably in TMDL programs and decrease the successes of implementations. The main purpose of this study is to establish a complete analysis framework to aid in address, analyze, and assess impact of uncertainties on TMDL programs. The importance of uncertainty analysis has been perceived but not completely applied. Uncertainty analysis approaches have been focus on quantifiable uncertainty effects, ignored the effects from non-quantifiable uncertainty elements. This leads to partial uncertainty analysis and might underestimate the results. Not only quantitative uncertainty but also qualitative (non-quantifiable) uncertainty should be considered in decision-making process. Due to the lack of a complete uncertainty analysis, especially for qualitative uncertainty, an integrated framework of uncertainty analysis with capable of evaluating both uncertainty effects is explored in this study. Qualitative uncertainty analysis (QLUA) with qualitative check list, belief function, and expert elicitation is developed to obtain the confidence level of target systems, which is TMDL program in this study. Quantitative uncertainty analysis (QTUA) is particularly designed for assessing the model parameter uncertainty on optimization programming of TMDL allocation. QTUA results in a distributed TMDL allocation and quantifies the uncertainty level (variability) of TMDL results. The both consequences are integrated into an uncertainty diagnose figure. The diagnose figure is able to manifest the uncertainty level of the system in terms of qualitative and quantitative uncertainty. If the uncertainty level is not complied with accepted level made by decision maker, the reliability of TMDL results might be questioned and uncertainty reduction is sought. As to uncertainty reduction, a new formulation, uncertainty value (UV), is created. The UV is used to represent the decision benefit obtained from every unit of uncertainty reduction, which is the confidence from QLUA and the variability from QTUA. Two case studies, KaoPing River and DaHan Creek, are used as applications of QLUA and QTUA, respectively. Although uncertainty exists in any decision steps and subjective assumptions (from expert elicitations) are unavoidable while seeking objective solutions, the comprehensive understandings are believed to improve quality of decision-making. The integrated framework of uncertainty analysis incorporating with individual evaluation approaches, diagnose figure, and uncertainty value function provides a systematic way to estimate uncertainty level in target systems, such as TMDL programs, and eventually; to increase the quality of decisions.